Between Concern and Epistemic Vigilance: How People Think about AI-generated Misinformation in Three Countries

This qualitative study examines how 45 digitally engaged news users in Mexico, the US, and the UK make sense of – and seek to navigate – AI-generated misinformation. Drawing on 90 semi-structured interviews and deliberative elicitation tasks conducted in 2024, we analyse participants’ own, often broad understandings of ‘misinformation’ and how these map onto concerns about synthetic text and, especially, politically salient AI-generated audio-visual content. Participants view generative AI as both a continuity (lowering the costs of producing misleading content) and rupture with the past due to lifelike synthetic audio-visual outputs that unsettle what counts as ‘real’ and the evidentiary status of images, audio, and video. However, rather than wholesale trust collapse, participants describe a brittle environment where credibility is judged relationally through source reputation, context, and provenance, creating a provenance premium, while ‘realism’ cues lose some of their value – a process we call indexical discounting. Risk assessments center on actors and institutions rather than artifacts, with trust adapted based on perceived institutional integrity. Participants acknowledge their own susceptibility to AI-generated content while viewing others as more vulnerable. They report tightening epistemic vigilance but recognise these practices are fragile and costly, demanding systemic, institutional, and product-level interventions. We discuss implications of our findings for the evolving ecology of trust and evidence. We suggest that generative AI may be closing a roughly 150–200-year ‘parenthesis’ in which recording technologies underwrote a distinctive form of evidential authority, returning credibility judgements to their older dependence on institutional and social trust.